Autonomous Driving With Perception Uncertainties: Deep-Ensemble Based Adaptive Cruise Control

Autonomous driving relies on perception systems to understand the environment and guide decision-making. Advanced perception systems often use black-box Deep Neural Networks (DNNs) for human-like comprehension, but their unpredictability and lack of interpretability can limit their use in safety-critical settings. This paper introduces an Ensemble of DNN regressors (Deep Ensemble) to provide predictions with quantified uncertainties. In the context of Adaptive Cruise Control (ACC), the Deep Ensemble estimates the distance headway to the lead vehicle from RGB images, allowing the downstream controller to account for uncertainty. An adaptive cruise controller is developed using Stochastic Model Predictive Control (MPC) with chance constraints to ensure probabilistic safety. Our ACC algorithm is evaluated using a high-fidelity traffic simulator and a real-world traffic dataset, demonstrating effective speed tracking and safe car following. The approach is also tested in out-of-distribution scenarios.